visual-rag-toolkit 0.1.4__tar.gz → 0.1.6__tar.gz

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: visual-rag-toolkit
-Version: 0.1.4
+Version: 0.1.6
 Summary: End-to-end visual document retrieval with ColPali, featuring two-stage pooling for scalable search
 Project-URL: Homepage, https://github.com/Ara-Yeroyan/visual-rag-toolkit
 Project-URL: Documentation, https://github.com/Ara-Yeroyan/visual-rag-toolkit#readme
@@ -94,6 +94,8 @@ End-to-end visual document retrieval toolkit featuring **fast multi-stage retrie
 
 **[Try the Live Demo](https://huggingface.co/spaces/Yeroyan/visual-rag-toolkit)** - Upload PDFs, index to Qdrant, and query with visual retrieval.
 
+**[Watch the Tutorial](https://drive.google.com/file/d/1SmpVJicsvyZ-awlwYtSkLw6BCSO7IJTw/view)** - Video walkthrough of the toolkit in action.
+
 This repo contains:
 - a **Python package** (`visual_rag`)
 - a **Streamlit demo app** (`demo/`)

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/README.md

@@ -9,6 +9,8 @@ End-to-end visual document retrieval toolkit featuring **fast multi-stage retrie
 
 **[Try the Live Demo](https://huggingface.co/spaces/Yeroyan/visual-rag-toolkit)** - Upload PDFs, index to Qdrant, and query with visual retrieval.
 
+**[Watch the Tutorial](https://drive.google.com/file/d/1SmpVJicsvyZ-awlwYtSkLw6BCSO7IJTw/view)** - Video walkthrough of the toolkit in action.
+
 This repo contains:
 - a **Python package** (`visual_rag`)
 - a **Streamlit demo app** (`demo/`)

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/demo/__init__.py

@@ -7,4 +7,4 @@ A Streamlit-based UI for:
 - Interactive playground for visual search
 """
 
-__version__ = "0.1.4"
+__version__ = "0.1.6"

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/demo/ui/header.py

@@ -26,5 +26,13 @@ def render_header():
         ">
             SIGIR 2026 Demo - Multi-Vector Visual Document Retrieval
         </p>
+        <p style="
+            font-size: 2rem;
+            margin-top: 8px;
+        ">
+            <a href="https://drive.google.com/file/d/1SmpVJicsvyZ-awlwYtSkLw6BCSO7IJTw/view" target="_blank" style="color: #0066cc; text-decoration: none;">
+                🎬 Watch Tutorial
+            </a>
+        </p>
     </div>
     """, unsafe_allow_html=True)

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "visual-rag-toolkit"
-version = "0.1.4"
+version = "0.1.6"
 description = "End-to-end visual document retrieval with ColPali, featuring two-stage pooling for scalable search"
 readme = "README.md"
 license = {file = "LICENSE"}
@@ -153,7 +153,7 @@ warn_unused_configs = true
 ignore_missing_imports = true
 
 [tool.bumpversion]
-current_version = "0.1.4"
+current_version = "0.1.6"
 parse = "(?P<major>\\d+)\\.(?P<minor>\\d+)\\.(?P<patch>\\d+)"
 serialize = ["{major}.{minor}.{patch}"]
 commit = true

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/__init__.py

@@ -33,7 +33,7 @@ Each component works independently - use only what you need.
 
 import logging
 
-__version__ = "0.1.4"
+__version__ = "0.1.6"
 
 
 def setup_logging(level: str = "INFO", format: str = None) -> None:

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/cli/main.py

@@ -288,7 +288,7 @@ def cmd_search(args):
         )
 
     # Search
-    query_np = query_embedding.detach().cpu().numpy()
+    query_np = query_embedding.detach().cpu().float().numpy()  # .float() for BFloat16
     if args.strategy == "single_full":
         results = single_stage.search(
             query_embedding=query_np,

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/retrieval/multi_vector.py

@@ -157,9 +157,10 @@ class MultiVectorRetriever:
     ) -> List[Dict[str, Any]]:
         q = self.embedder.embed_query(query)
         if isinstance(q, torch.Tensor):
-            query_embedding = q.detach().cpu().numpy()
+            # .float() converts BFloat16 to Float32 (numpy doesn't support BFloat16)
+            query_embedding = q.detach().cpu().float().numpy()
         else:
-            query_embedding = np.asarray(q)
+            query_embedding = np.asarray(q, dtype=np.float32)
 
         return self.search_embedded(
             query_embedding=query_embedding,

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/retrieval/single_stage.py

@@ -129,5 +129,5 @@ class SingleStageRetriever:
         if isinstance(embedding, torch.Tensor):
             if embedding.dtype == torch.bfloat16:
                 return embedding.cpu().float().numpy()
-            return embedding.cpu().numpy()
+            return embedding.cpu().float().numpy()  # .float() for BFloat16 compatibility
         return np.array(embedding, dtype=np.float32)

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/retrieval/three_stage.py

@@ -51,7 +51,7 @@ class ThreeStageRetriever:
         if isinstance(embedding, torch.Tensor):
             if embedding.dtype == torch.bfloat16:
                 return embedding.cpu().float().numpy()
-            return embedding.cpu().numpy()
+            return embedding.cpu().float().numpy()  # .float() for BFloat16 compatibility
         return np.array(embedding, dtype=np.float32)
 
     def _infer_vector_is_multivector(self, vector_name: str) -> bool:

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/retrieval/two_stage.py

@@ -418,7 +418,7 @@ class TwoStageRetriever:
         if isinstance(embedding, torch.Tensor):
             if embedding.dtype == torch.bfloat16:
                 return embedding.cpu().float().numpy()
-            return embedding.cpu().numpy()
+            return embedding.cpu().float().numpy()  # .float() for BFloat16 compatibility
         return np.array(embedding, dtype=np.float32)
 
     def build_filter(

{visual_rag_toolkit-0.1.4 → visual_rag_toolkit-0.1.6}/visual_rag/visualization/saliency.py

@@ -55,14 +55,14 @@ def generate_saliency_map(
     if hasattr(query_embedding, "numpy"):
         query_np = query_embedding.numpy()
     elif hasattr(query_embedding, "cpu"):
-        query_np = query_embedding.cpu().numpy()
+        query_np = query_embedding.cpu().float().numpy()  # .float() for BFloat16
     else:
         query_np = np.array(query_embedding, dtype=np.float32)
 
     if hasattr(doc_embedding, "numpy"):
         doc_np = doc_embedding.numpy()
     elif hasattr(doc_embedding, "cpu"):
-        doc_np = doc_embedding.cpu().numpy()
+        doc_np = doc_embedding.cpu().float().numpy()  # .float() for BFloat16
     else:
         doc_np = np.array(doc_embedding, dtype=np.float32)